Conversion rate optimization (CRO) is the practice of systematically improving the percentage of users who complete a desired action: signing up, purchasing, upgrading, completing onboarding, or any other step that matters to the business. The "systematic" part is what separates CRO from guessing. It means forming hypotheses about why users drop off, testing changes in controlled experiments, and making decisions based on causal evidence.
CRO has a reputation as a marketing discipline focused on landing pages and button colors. That reputation undersells it. The same methodology applies to any product flow where users move through a sequence of steps and some percentage drops off at each one. Checkout funnels, onboarding sequences, upgrade paths, search-to-play journeys: they're all conversion problems, and they all respond to the same experimental approach.
How does CRO connect to A/B testing?
A/B testing is the primary tool in CRO. You identify a conversion step where drop-off is high, hypothesize why users are leaving, build a variant that addresses the hypothesis, and split traffic between the original and the variant. If the variant produces a statistically significant improvement in conversion rate, you've found something worth shipping.
The rigor matters. CRO without controlled experiments is pattern matching: you see a high drop-off rate, you redesign the page, conversion goes up (or down), and you don't know whether the change caused the shift or whether something else happened at the same time. A/B testing isolates the causal effect of your specific change.
The statistical requirements for CRO experiments are often demanding. Conversion rates are binary metrics (the user either converted or didn't), which tend to have high variance relative to the effect sizes that matter. A 0.5 percentage point improvement in a 3% conversion rate is a meaningful 17% relative lift, but detecting it requires substantial sample size and adequate statistical power. Underpowered CRO tests are a common failure mode: the team runs the test for a week, sees no significance, and concludes "it didn't work," when the test simply didn't have enough data to detect a real effect.
What makes CRO experiments different from feature experiments?
CRO experiments typically target a specific, measurable business outcome (conversion rate at a defined step) rather than a broader product metric (engagement, retention). This focus makes them easier to design but creates a specific risk: optimizing a conversion step in isolation can hurt the broader user experience.
A classic example: making a "Sign Up" modal harder to dismiss increases signup conversion but decreases long-term retention because users who were coerced into signing up weren't genuinely interested. The conversion metric improved. The business outcome didn't.
This is why multi-metric decision making matters in CRO. Every CRO experiment should have guardrail metrics that monitor downstream effects. Did the conversion improvement come at the cost of increased churn, lower engagement after conversion, or higher support volume? Confidence structures experiments around this distinction, separating success metrics (what you're optimizing) from guardrail metrics (what you're protecting).
Where does CRO fit in the product loop?
CRO works best as part of a continuous discovery process, not as a one-time optimization project. User behavior changes, the product changes, and what worked six months ago may not work today. The most effective CRO programs run ongoing experiments on conversion-critical flows, treating optimization as a permanent practice rather than a seasonal initiative.
At mature experimentation organizations, CRO experiments share infrastructure with feature experiments. The same platform, the same statistical methods, the same metric pipelines. Confidence handles both use cases: a CRO experiment optimizing a signup flow and a feature experiment testing a new recommendation algorithm use the same flag evaluation, the same warehouse-native analysis, and the same guardrail monitoring.